基于遗传算法与改进与GM (1, 1) 模型IGM (1, 1) 直接建模.docVIP

基于遗传算法的改进的GM(1,1)模型IGM(1,1)直接建模 郑照宁，刘德顺 (清华大学现代管理研究中心，清华大学技术经济与能源系统分析研究所，北京，100084) 摘 要 GM(1,1)模型一般以模型还原值与实际值平均相对误差检验模型的模拟精度.本文以模型还原值与实际值平均相对误差最小化为目标函数将GM(1,1)模型转化成一个不用进行灰微分方程参数辨识的优化模型，我们称之为改进的GM(1,1)模型,简称IGM(1,1)。IGM(1,1)避开了灰微分方程参数辨识时的合理选取背景值的问题，实现GM(1,1)模型的直接建模。由于IGM(1,1)目标函数非连续，不可导，用传统的优化无法求解，本文针对IGM(1,1)模型的特性设计了求解该优化模型的遗传算法并进行了算例验证，求解结果表明了IGM(1,1)模型的模拟精度远高于GM(1,1)模型。 关键词 GM(1,1)；改进的GM(1,1)模型IGM(1,1); 背景值；遗传算法 Direct Modeling Improved GM (1,1) Model IGM (1,1) by Genetic Algorithm ZHENG Zhao-ning LIU De-shun (Modern Management Research Center of Tsinghua University, Institute of Techno-Economics and Energy System Analysis of Tsinghua University, Beijing, 100084, P.R. China) Abstract: Generally GM (1,1) model takes the average relative error between restored value of the model and real value as the criterion to evaluate the simulation precision. In this paper, GM (1,1) model was converted to an optimization model, which doesn’t need to identify the parameters of grey differential equation, using the average relative error between restored value of the model and real value as objective function. The model was called Improve GM (1,1) model, IGM (1,1) for short. IGM (1,1) avoids the problem how to rationally select background values in parameter identification of grey differential equation and realize the direct modeling of GM (1,1). The object function of IGM(1,1) is unable to be gained by classical optimization approaches due to its discontinuousness and non-differentiability. We design a genetic algorithm for IGM(1,1) based on its characteristics and test the algorithm with an example. The result acquired shows that the simulation precision of IGM(1,1) model is much higher than that of GM(1,1) mode Key words: GM(1,1), improved GM(1,1) model IGM(1,1), background value, genetic algorithm 1.改进的GM(1,1)模型IGM(1,1) 自从80年代中期邓聚龙教授提出GM(1,1)模型以来,由于只需少量数据便可建模,得到了广泛的应用[1].GM(1,1)是建立在灰导数和基础上的灰微分模型,其背景值一般由相邻累加数平均生成.由于相邻累加数之间是空的,用均值 (1) 来代表灰微分方程